biLGP: The bivariate Lagrangian Poisson (LGP) distribution

In RMKdiscrete: Sundry Discrete Probability Distributions

The bivariate Lagrangian Poisson (LGP) distribution

Description

Density, random-number generation, and moments of the log-transformed distribution.

Usage

dbiLGP(y, theta, lambda, nc=NULL, log=FALSE, add.carefully=FALSE)
biLGP.logMV(theta,lambda,nc=NULL,const.add=1,tol=1e-14,add.carefully=FALSE)
rbiLGP(n, theta, lambda)

Arguments

y	Numeric vector or two-column matrix of bivariate data. If matrix, each row corresponds to an observation.
theta	Numeric vector or three-column matrix of non-negative values for index parameters theta0, theta1, and theta2, in that order. If matrix, is read by row.
lambda	Numeric vector or three-column matrix of values for multiplicative parameters lambda0, lambda1, and lambda2, in that order. If matrix, is read by row. Values must be on the interval [-1,1].
nc	Numeric vector or three-column matrix of (reciprocals of) the normalizing constants. These constants differ from 1 only if the corresponding lambda parameter is negative; see dLGP() for details. If matrix, is read by row. Defaults to NULL, in which case the normalizing constants are computed automatically.
log	Logical; should the natural log of the probability be returned? Defaults to FALSE.
add.carefully	Logical. If TRUE, the program takes extra steps to try to prevent round-off error during the addition of probabilities. Defaults to FALSE, which is recommended, since using TRUE is slower and rarely makes a noticeable difference in practice.
const.add	Numeric vector of positive constants to add to the non-negative integers before taking their natural logarithm. Defaults to 1, for the typical log(y+1) transformation.
tol	Numeric; must be positive. When biLGP.logMV() is calculating the second moment of the log-transformed distribution, it stops when the next term in the series is smaller than tol.
n	Integer; number of observations to be randomly generated.

Details

The bivariate LGP is constructed from three independent latent random variables, X0, X1, and X2, where

X0 ~ LGP(theta0, lambda0)

X1 ~ LGP(theta1, lambda1)

X2 ~ LGP(theta2, lambda2)

The observable variables, Y1 and Y2, are defined as Y1 = X0 + X1 and Y2 = X0 + X2, and thus the dependence between Y1 and Y2 arises because of the common term X0. The joint PMF of Y1 and Y2 is derived from the joint PMF of the three independent latent variables, with X1 and X2 re-expressed as Y1 - X0 and Y2 - X0, and after X0 is marginalized out.

Function dbiLGP() is the bivariate LGP density (PMF). Function rbiLGP() generates random draws from the bivariate LGP distribution, via calls to rLGP(). Function biLGP.logMV() numerically computes the means, variances, and covariance of a bivariate LGP distribution, after it has been log transformed following addition of a positive constant.

Vectors of numeric arguments other than tol are cycled, whereas only the first element of logical and integer arguments is used.

Value

dbiLGP() returns a numeric vector of probabilities. rbiLGP() returns a matrix of random draws, which is of type 'numeric' (rather than 'integer', even though the bivariate LGP only has support on the non-negative integers). biLGP.logMV() returns a numeric matrix with the following five named columns:

1. EY1: Post-tranformation expectation of Y1.

2. EY2: Post-tranformation expectation of Y2.

3. VY1: Post-tranformation variance of Y1.

4. VY2: Post-tranformation variance of Y2.

5. COV: Post-tranformation covariance of Y1 and Y2.

Author(s)

Robert M. Kirkpatrick [e-mail]

References

Famoye, F., & Consul, P. C. (1995). Bivariate generalized Poisson distribution with some applications. Metrika, 42, 127-138.

Consul, P. C., & Famoye, F. (2006). Lagrangian Probability Distributions. Boston: Birkhauser.

See Also

LGP, dpois()

Examples

## The following two lines do the same thing:
dbiLGP(y=1,theta=1,lambda=0.1)
dbiLGP(y=c(1,1),theta=c(1,1,1),lambda=c(0.1,0.1,0.1))

dbiLGP(y=c(1,1,2,2,3,5),theta=c(1,1,1,2,2,2),lambda=0.1)
## Due to argument cycling, the above line is doing the following three steps:
dbiLGP(y=c(1,1),theta=c(1,1,1),lambda=c(0.1,0.1,0.1))
dbiLGP(y=c(2,2),theta=c(2,2,2),lambda=c(0.1,0.1,0.1))
dbiLGP(y=c(3,5),theta=c(1,1,1),lambda=c(0.1,0.1,0.1))

## Inputs to dbiLGP() can be matrices, too:
dbiLGP(y=matrix(c(1,1,2,2,3,5),ncol=2,byrow=TRUE),
  theta=matrix(c(1,1,1,2,2,2,1,1,1),ncol=3,byrow=TRUE),
  lambda=0.1)

## theta0 = 0 implies independence:
a <- dbiLGP(y=c(1,3),theta=c(0,1,2),lambda=c(0.1,-0.1,0.5))
b <- dLGP(x=1,theta=1,lambda=-0.1) * dLGP(x=3,theta=2,lambda=0.5)
a-b #<--near zero.
## lambdas of zero yield the ordinary Poisson:
a <- dbiLGP(y=c(1,3), theta=c(0,1,2),lambda=0)
b <- dpois(x=1,lambda=1) * dpois(x=3,lambda=2) #<--LGP theta is pois lambda
a-b #<--near zero

( y <- rbiLGP(10,theta=c(1.1,0.87,5.5),lambda=c(0.87,0.89,0.90)) )
dbiLGP(y=y,theta=c(1.1,0.87,5.5),lambda=c(0.87,0.89,0.90))

biLGP.logMV(theta=c(0.65,0.35,0.35),lambda=0.7,tol=1e-8)

RMKdiscrete documentation built on May 8, 2022, 1:05 a.m.